TFE: Adds a destructor to ResourceVariables in Python that destroys the underlying resource. This makes the lifetime of the underlying resource match that of its corresponding Python object.

PiperOrigin-RevId: 172039259

1 files changed, 42 insertions, 40 deletions

diff --git a/tensorflow/python/training/adam_test.py b/tensorflow/python/training/adam_test.py
index defcf33714..96de9b921b 100644
--- a/tensorflow/python/training/adam_test.py
+++ b/tensorflow/python/training/adam_test.py
@@ -29,6 +29,7 @@ from tensorflow.python.framework import test_util
 from tensorflow.python.ops import array_ops
 from tensorflow.python.ops import math_ops
 from tensorflow.python.ops import resource_variable_ops
+from tensorflow.python.ops import variable_scope
 from tensorflow.python.ops import variables
 from tensorflow.python.platform import test
 from tensorflow.python.training import adam
@@ -152,53 +153,54 @@ class AdamOptimizerTest(test.TestCase):
 
   def doTestBasic(self, use_resource=False):
     for i, dtype in enumerate([dtypes.half, dtypes.float32, dtypes.float64]):
-      # Initialize variables for numpy implementation.
-      m0, v0, m1, v1 = 0.0, 0.0, 0.0, 0.0
-      var0_np = np.array([1.0, 2.0], dtype=dtype.as_numpy_dtype)
-      grads0_np = np.array([0.1, 0.1], dtype=dtype.as_numpy_dtype)
-      var1_np = np.array([3.0, 4.0], dtype=dtype.as_numpy_dtype)
-      grads1_np = np.array([0.01, 0.01], dtype=dtype.as_numpy_dtype)
-
-      if use_resource:
-        var0 = resource_variable_ops.ResourceVariable(
-            var0_np, name="var0_%d" % i)
-        var1 = resource_variable_ops.ResourceVariable(
-            var1_np, name="var1_%d" % i)
-      else:
-        var0 = variables.Variable(var0_np)
-        var1 = variables.Variable(var1_np)
-      grads0 = constant_op.constant(grads0_np)
-      grads1 = constant_op.constant(grads1_np)
-
-      opt = adam.AdamOptimizer()
-      update = opt.apply_gradients(zip([grads0, grads1], [var0, var1]))
+      with variable_scope.variable_scope("%d" % i):
+        # Initialize variables for numpy implementation.
+        m0, v0, m1, v1 = 0.0, 0.0, 0.0, 0.0
+        var0_np = np.array([1.0, 2.0], dtype=dtype.as_numpy_dtype)
+        grads0_np = np.array([0.1, 0.1], dtype=dtype.as_numpy_dtype)
+        var1_np = np.array([3.0, 4.0], dtype=dtype.as_numpy_dtype)
+        grads1_np = np.array([0.01, 0.01], dtype=dtype.as_numpy_dtype)
 
-      if context.in_graph_mode():
-        self.evaluate(variables.global_variables_initializer())
-        # Fetch params to validate initial values
-        self.assertAllClose([1.0, 2.0], self.evaluate(var0))
-        self.assertAllClose([3.0, 4.0], self.evaluate(var1))
+        if use_resource:
+          var0 = resource_variable_ops.ResourceVariable(
+              var0_np, name="var0_%d" % i)
+          var1 = resource_variable_ops.ResourceVariable(
+              var1_np, name="var1_%d" % i)
+        else:
+          var0 = variables.Variable(var0_np)
+          var1 = variables.Variable(var1_np)
+        grads0 = constant_op.constant(grads0_np)
+        grads1 = constant_op.constant(grads1_np)
 
-      beta1_power, beta2_power = opt._get_beta_accumulators()
+        opt = adam.AdamOptimizer()
+        update = opt.apply_gradients(zip([grads0, grads1], [var0, var1]))
 
-      # Run 3 steps of Adam
-      for t in range(1, 4):
         if context.in_graph_mode():
-          self.evaluate(update)
-        elif t > 1:
-          opt.apply_gradients(zip([grads0, grads1], [var0, var1]))
+          self.evaluate(variables.global_variables_initializer())
+          # Fetch params to validate initial values
+          self.assertAllClose([1.0, 2.0], self.evaluate(var0))
+          self.assertAllClose([3.0, 4.0], self.evaluate(var1))
 
-        self.assertAllCloseAccordingToType(0.9**(t + 1),
-                                           self.evaluate(beta1_power))
-        self.assertAllCloseAccordingToType(0.999**(t + 1),
-                                           self.evaluate(beta2_power))
+        beta1_power, beta2_power = opt._get_beta_accumulators()
 
-        var0_np, m0, v0 = adam_update_numpy(var0_np, grads0_np, t, m0, v0)
-        var1_np, m1, v1 = adam_update_numpy(var1_np, grads1_np, t, m1, v1)
+        # Run 3 steps of Adam
+        for t in range(1, 4):
+          if context.in_graph_mode():
+            self.evaluate(update)
+          elif t > 1:
+            opt.apply_gradients(zip([grads0, grads1], [var0, var1]))
+
+          self.assertAllCloseAccordingToType(0.9**(t + 1),
+                                             self.evaluate(beta1_power))
+          self.assertAllCloseAccordingToType(0.999**(t + 1),
+                                             self.evaluate(beta2_power))
+
+          var0_np, m0, v0 = adam_update_numpy(var0_np, grads0_np, t, m0, v0)
+          var1_np, m1, v1 = adam_update_numpy(var1_np, grads1_np, t, m1, v1)
 
-        # Validate updated params
-        self.assertAllCloseAccordingToType(var0_np, self.evaluate(var0))
-        self.assertAllCloseAccordingToType(var1_np, self.evaluate(var1))
+          # Validate updated params
+          self.assertAllCloseAccordingToType(var0_np, self.evaluate(var0))
+          self.assertAllCloseAccordingToType(var1_np, self.evaluate(var1))
 
   def testBasic(self):
     with self.test_session():